R help - Nov 2009 - wilcox.test loop through variable names

Often I perform the same task on a series of variables in a dataframe,
by looping through a character vector that holds the names and using
paste(), eval(), and parse() inside the loop.

For instance:

rm(environmental)
thesevars<-names(environmental)
environmental$ToyReal <-rnorm(nrow(environmental)) 
environmental$ToyDichot<- environmental$ToyReal < 0.53

tableOfResults<-data.frame(var=thesevars)

tableOfResults$p_wilcox <- NA

tableOfResults$Beta_lm <- NA

rownames(tableOfResults)<-thesevars

for( thisvar in thesevars) {
  	thiscommand<- paste("thiswilcox <- wilcox.test (", thisvar,
" ~ ToyDichot , data=environmental)")
 	eval(parse(text=thiscommand))
  	tableOfResults[thisvar, "p_wilcox"] <- thiswilcox$p.value
 	thislm<-lm( environmental[ c( "ToyReal", thisvar )])
  	tableOfResults[thisvar, "Beta_lm"] <- coef(thislm)[thisvar]
}

print(tableOfResults)

Of course, the loop above is a toy example. In real life I might first figure
out whether the variable is
continuous, dichotomous, or categorical taking on several values, then perform
an operation depending on
its type.

The use of paste(), eval(), and parse() seems awkward.  As Gabor Grothendieck
showed
(http://tolstoy.newcastle.edu.au/R/e8/help/09/11/4520.html), if we
are calling a regression function such as lm() we can avoid using
paste(), as shown above.

But is there a way to avoid paste() and eval() when one uses t.test()
or wilcox.test()?

Thanks

Jacob A. Wegelin
Department of Biostatistics
Virginia Commonwealth University
Richmond VA 23298-0032
U.S.A.

On Sun, 15 Nov 2009 14:33 -0500, "Jacob Wegelin"
<jacobwegelin at fastmail.fm> wrote:
> 
> Often I perform the same task on a series of variables in a dataframe,
> by looping through a character vector that holds the names and using
> paste(), eval(), and parse() inside the loop.
> 
> For instance:
> 
> rm(environmental)
> thesevars<-names(environmental)
> environmental$ToyReal <-rnorm(nrow(environmental)) 
> environmental$ToyDichot<- environmental$ToyReal < 0.53
> 
> tableOfResults<-data.frame(var=thesevars)
> 
> tableOfResults$p_wilcox <- NA
> 
> tableOfResults$Beta_lm <- NA
> 
> rownames(tableOfResults)<-thesevars
> 
> for( thisvar in thesevars) {
>   	thiscommand<- paste("thiswilcox <- wilcox.test (",
thisvar, " ~ ToyDichot , data=environmental)")
>  	eval(parse(text=thiscommand))
>   	tableOfResults[thisvar, "p_wilcox"] <- thiswilcox$p.value
>  	thislm<-lm( environmental[ c( "ToyReal", thisvar )])
>   	tableOfResults[thisvar, "Beta_lm"] <- coef(thislm)[thisvar]
> }
> 
> print(tableOfResults)
> 
> Of course, the loop above is a toy example. In real life I might first
> figure out whether the variable is
> continuous, dichotomous, or categorical taking on several values, then
> perform an operation depending on
> its type.
> 
> The use of paste(), eval(), and parse() seems awkward.  As Gabor
> Grothendieck showed
> (http://tolstoy.newcastle.edu.au/R/e8/help/09/11/4520.html), if we
> are calling a regression function such as lm() we can avoid using
> paste(), as shown above.
> 
> But is there a way to avoid paste() and eval() when one uses t.test()
> or wilcox.test()?
Here is a solution:

rm(environmental)
thesevars<-names(environmental)
environmental$ToyReal <-rnorm(nrow(environmental))
environmental$ToyDichot<- environmental$ToyReal < 0.53

ThisList<-
lapply( environmental[thesevars], function( OneVar ) {
   c(
      p_wilcox= wilcox.test( OneVar ~ environmental$ToyDichot )$p.value
         ,
      Beta_lm = as.numeric(coef(lm( environmental$ToyReal ~ OneVar
      ))["OneVar"])
   )
   }
)

do.call("rbind", ThisList)

Jacob A. Wegelin 
Department of Biostatistics 
Virginia Commonwealth University 
Richmond VA 23298-0032 
U.S.A.